> For the complete documentation index, see [llms.txt](https://docs.biobox.io/guide/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.biobox.io/guide/how-to/ask-biobox-with-claude.md).

# Use the MCP to connect your agents to BioBox

The **BioBox MCP** connects an AI assistant —  on the web, in the desktop app, or in terminal — directly to your BioBox knowledge graph and all of the reasoning systems you have configured on the platform. Once connected, you can ask questions in plain language and the agent answers using **your organization's data**, not the open internet.

Under the hood, it uses the [Model Context Protocol (MCP)](https://modelcontextprotocol.io), an open standard that lets an assistant call a defined set of tools. BioBox exposes your graph as those tools, so the agent can search entities, read prioritization reports, pull evidence, and trace relationships — and then explain the result back to you with the numbers it actually retrieved.

## Reasoning Infrastructure for your Agents

The purpose of the MCP is to provide agents with the tools necessary to reason like the scientists in your organization while ensuring your institutional knowledge remains modular - ready to plug into any agentic system you choose, and more importantly, remains sovereign. Details on setting up your first [Reasoning System can be found here](/guide/how-to/create-prioritization-graph-models.md). 

To get the most out of the MCP we recommend setting up a few Reasoning Systems on the platform. These are modules that reflect how your teams make decisions and connect directly to the data in your knowledge graph to rank and prioritize data points. 

**The types of questions you can ask depend on the data you have available in your knowledge graph and how you have configured your graph schema.** 

The MCP is built for the questions a scientist asks when making portfolio evaluation decisions, for example:

* **Target prioritization** — *"Across my reasoning systems, what are the top 10 targets suitable for ALS and atopic dermatitis?"* 
* **Understanding how & why decisions changed** - *"Why did SOD1 get deprioritized for our ALS program? What data and evidence changed?"*
* **Evidence & provenance** — *"What is all of the evidence to support kerotonicytes as an important cell population for psoriasis?"*
* **Multi-modal questions** — *"What are genes that are upregulated in ALS macrophage cell populations and have a genetic association to the disease?"*
* **Understanding your reason systems** — *"Which reasoning systems are available, and what does each prioritize?"*
* **Looking things up** — *"What entity types are in the graph?"* or *"What are all of the  psoriasis datasets in my graph?" or "What are all the drug targets in phase 1 ALS clinical trials?"*

Every answer is grounded: the agent names the report, score, or path it used, so you can trace it back into the BioBox app. 

## What it can and can't do

{% hint style="info" %}
The BioBox MCP is **read-only** and **organization-scoped**.
{% endhint %}

* **Read-only.** It can search, score, rank, and explain — it **cannot** modify your graph, edit reports, change models, or delete anything. Ask it anything; it can't break anything.
* **Scoped to your organization.** You see exactly the data your BioBox account already has access to. Signing in uses your own BioBox identity, and every request is checked against the same permissions as the app. The MCP **invents no new access** — if you can't see it in BioBox, Claude can't see it either.
* **Grounded, not generative about your data.** Facts about entities, scores, and rankings come from the graph via defined tools. Claude does the language and reasoning; BioBox provides the numbers.

## Get started

1. [**Connect Claude (sign in)**](/guide/how-to/ask-biobox-with-claude/connect-claude.md) — add the connector in Claude web, Claude Code, or Claude Desktop and sign in with your BioBox account.
2. [**Example Question Workflow**](/guide/how-to/ask-biobox-with-claude/quickstart.md) — An example of some of the questions you can ask. 
3. [**Tool Reference**](/guide/how-to/ask-biobox-with-claude/tools.md) — the full set of tools Claude can call, with parameters and example responses.